What is the proper way to stretch an image?

[georgian82]

Hi guys,

I have been having some trouble with the most current version of PI, specifically with the way the image looks after integration. I have been playing around with every possible setting during calibration and integration to make sure I am not doing something wrong and I came up across an interesting finding which made me question the way I have been stretching my data post integration (and it could also explain the issues I have been having).

Basically, when I was applying the auto stretch to the image with the "linkage" option enabled, it just looked wrong. Almost like if it had a film on top of it and the pixels got distorted and the image is impossible to edit. On the other hand, when applying auto stretch but with the "linkage" option disabled, the image looks the way I would expect. What is most interesting is that with the "linkage" option disabled, I don't even have to apply DBE like I do when the option is enabled.

Until now, I have always applied the auto stretch relying on the default settings. But now I do want to learn more about it and I hope that somebody can guide me in the right direction as I feel is an important step.

I have uploaded two images in Dropbox. One is with the auto stretch icon with "linkage" option enabled and the second is with the "linkage" option disabled for reference.

https://www.dropbox.com/s/d5rc49tk1n6r0x7/Elephant%20Trunk%20Nebula%20-%20Integration%20-%20stretched%20no%20link%202.xisf?dl=0

https://www.dropbox.com/s/ttozog3rco9al2t/Elephant%20Turnk%20Nebula%20-%20stretched%20linked%202.xisf?dl=0

Thank you all!

Sebastian

[pscammp]

Sebastian,
     No expert here but from what im seeing you have not used the correct bayer pattern to debayer your Lights while preprocessing.

What Camera was used for this imaging session   

Paul

[georgian82]

Sebastian,
     No expert here but from what im seeing you have not used the correct bayer pattern to debayer your Lights while preprocessing.

What Camera was used for this imaging session   

Paul

Hi Paul,

Thanks for the reply. I used the automatic setting at the debayeing step which as I understand works well. I am using an Olympus EPM2 micro four thirds camera.

Thanks
Sebastian

[pscammp]

As a general rule once you have your final integration, unless you are dealing with mono narrow band images, you will always have to have the 'Screen Transfer Function' set to unlink the channels, this is because all (or most anyway) camera's will have some kind of 'Colour Bias' caused by the specific camera's bayer system. Mine for instance just gives me an almost completely red image if channels are linked. My old Canon 1000D gave me a green image when linked. Applying Background Neutralization to my images takes out 90% of this colour bias but it's not until I perform 'Colour Calibration' that all the colour bias is finally gone and I can continue with the colour channels linked.

Hope This Helps !

Paul

[pscammp]

Seb,
   According to official specs for your camera your debayering pattern is: RGBG

On the following web site:

http://www.imaging-resource.com/PRODS/olympus-epm2/olympus-epm2DAT.HTM

It shows next to 'Colour Filter Type'

Maybe the 'Auto' setting is not picking the correct debayer pattern of RGBG

Try processing again but physically pick the correct pattern instead of leaving it to Auto

Try it on a single sub and see how it looks

Good Luck
Paul

[Luca Amerio]

Not an expert here, but if you haven't performed the color-calibration is normal that the image colors are distorted.

Anyway, that's just the screen transfer function, that is a "temporary stretch" just to visualize the data. You aren't really affecting your image.

To do so you must use either the Histogram Transformation or the Masked Stretch. But before doing that, you need to color-calibrate the channels...

[georgian82]

As a general rule once you have your final integration, unless you are dealing with mono narrow band images, you will always have to have the 'Screen Transfer Function' set to unlink the channels, this is because all (or most anyway) camera's will have some kind of 'Colour Bias' caused by the specific camera's bayer system. Mine for instance just gives me an almost completely red image if channels are linked. My old Canon 1000D gave me a green image when linked. Applying Background Neutralization to my images takes out 90% of this colour bias but it's not until I perform 'Colour Calibration' that all the colour bias is finally gone and I can continue with the colour channels linked.

Hope This Helps !

Paul

That is very helpful to know. That explains why I get a completely green image after applying screen transfer function with the channels linked. I used to simply apply DBE to remove it followed by color calibration and it worked well. But for some reason I cannot do that anymore since after applying DBE there is almost like a clear film over the image that doesn't allow me to edit the data properly (which is an issue I didn't encounter before).

I guess I will just stick to applying the screen transfer function with unlinked channels going forward.

Thanks a lot!

[georgian82]

Not an expert here, but if you haven't performed the color-calibration is normal that the image colors are distorted.

Anyway, that's just the screen transfer function, that is a "temporary stretch" just to visualize the data. You aren't really affecting your image.

To do so you must use either the Histogram Transformation or the Masked Stretch. But before doing that, you need to color-calibrate the channels...

Hi Luca,

Thanks for the reply. I applied local normalization so I figured the color calibration would not be necessary?

Yes, I understand that I need to permanently save the stretched image by using the histogram, but I am really more asking about the previous step, stretching the data properly first in order to get the best representation possible.

Thanks again,

Sebastian

[georgian82]

Seb,
   According to official specs for your camera your debayering pattern is: RGBG

On the following web site:

http://www.imaging-resource.com/PRODS/olympus-epm2/olympus-epm2DAT.HTM

It shows next to 'Colour Filter Type'

Maybe the 'Auto' setting is not picking the correct debayer pattern of RGBG

Try processing again but physically pick the correct pattern instead of leaving it to Auto

Try it on a single sub and see how it looks

Good Luck
Paul

Hi Paul,

Thanks for the info! I will give it a try manually (with the correct bayer pattern) and see if I see a difference from the auto setting.

Cheers!

[pscammp]

Not an expert here, but if you haven't performed the color-calibration is normal that the image colors are distorted.

Anyway, that's just the screen transfer function, that is a "temporary stretch" just to visualize the data. You aren't really affecting your image.

To do so you must use either the Histogram Transformation or the Masked Stretch. But before doing that, you need to color-calibrate the channels...

Hi Luca,

Thanks for the reply. I applied local normalization so I figured the color calibration would not be necessary?

Yes, I understand that I need to permanently save the stretched image by using the histogram, but I am really more asking about the previous step, stretching the data properly first in order to get the best representation possible.

Thanks again,

Sebastian

Seb,
   Local Normalization is a new PI process not even slightly related to colour correction, using it does not mean you don't need to do colour correction, colour correction is an essential step in your workflow. Have a read through the following thread by Juan, he talks about the 'Frame Adaption' process -- This is actually the Local Normalization process we see today with it's final name.

https://pixinsight.com/forum/index.php?topic=11063.0

Regards
Paul

[georgian82]

Not an expert here, but if you haven't performed the color-calibration is normal that the image colors are distorted.

Anyway, that's just the screen transfer function, that is a "temporary stretch" just to visualize the data. You aren't really affecting your image.

To do so you must use either the Histogram Transformation or the Masked Stretch. But before doing that, you need to color-calibrate the channels...

Hi Luca,

Thanks for the reply. I applied local normalization so I figured the color calibration would not be necessary?

Yes, I understand that I need to permanently save the stretched image by using the histogram, but I am really more asking about the previous step, stretching the data properly first in order to get the best representation possible.

Thanks again,

Sebastian

Seb,
   Local Normalization is a new PI process not even slightly related to colour correction, using it does not mean you don't need to do colour correction, colour correction is an essential step in your workflow. Have a read through the following thread by Juan, he talks about the 'Frame Adaption' process -- This is actually the Local Normalization process we see today with it's final name.

https://pixinsight.com/forum/index.php?topic=11063.0

Regards
Paul

Paul,

Thank you very much for the link. Obviously the process of local normalization is new to me and I erroneously thought it had something to do with color correction.

Thanks again...I appreciate the feedback.

Sebastian

[the Elf]

A simple method to determine the bayer pattern is to put some colored stuff on the table and take a raw photo of it. Load it in PI and try the patterns until it shows up with natural colors. I did with water and soft drink bottles with blue, green, red and yellow lables. The result is very obvious. With a deep sky object it is not that easy.